Founders interview users every week, then lose the synthesis when the next sprint starts. The pattern that works: Grain records and transcribes, Dovetail holds the tagged clips, Notion holds the founder-facing memo, and a research agent stitches them into one workflow. The founder reviews insights one row at a time. The memo stays attributed, so a new teammate or a future agent can trace any claim back to a real call. For the wider pillar, see Dock for founders.
Notion, Grain, and Dovetail stay the system of record for the raw data. Dock is the system of record for what the AGENT INTERPRETS. Each Dock row carries a pointer back to the platform record, agent identity, decision, reviewer, and timestamp. The agent re-fetches platform data via fresh API reads when it needs current state.
The CustomerInsights surface
| insight_id | source_call (Grain) | dovetail_tag | claim | confidence | reviewer | status |
|---|---|---|---|---|---|---|
| INS-204 | grain.com/call/9f12 (Acme, 2026-05-22) | onboarding-friction | "Users abandon setup at the integrations step because OAuth scopes are unclear." | 0.82 | founder@ | approved |
| INS-205 | grain.com/call/a730 (Pylon, 2026-05-24) | pricing-confusion | "Self-serve buyers expect per-seat pricing and bounce on usage-based." | 0.71 | founder@ | needs-followup |
| INS-206 | grain.com/call/b855 (Replo, 2026-05-25) | feature-request-export | "Three of five teams asked for a CSV export of agent decisions." | 0.90 | founder@ | approved |
Each row is one synthesized claim. The Grain link is the citation. The Dovetail tag is the cluster. The founder either approves the row, marks it for follow-up, or rejects it. Nothing reaches the memo without a status of approved.
The workflow
A founder finishes a 30-minute call with a design partner. Grain transcribes it within five minutes. The research agent reads the transcript, pulls candidate quotes, and proposes Dovetail tags from the existing taxonomy. It then drafts one to three rows in CustomerInsights, each pointing back to the Grain timestamp and the Dovetail clip. The founder opens Dock, reads the proposed claim, listens to the 20-second clip, and approves. Approved rows append to a Notion memo titled "What we learned this week," grouped by tag. Every claim in the memo carries a footnote linking to the source call. The agent never edits the memo's narrative paragraphs. It only appends new evidence rows under each theme. This division (agent for evidence, founder for narrative) is the same separation covered in the agent collaboration primer and applied to the research function in Dock for research.
Why it matters
Founders forget which claim came from which call by Friday. Tagging clips in Dovetail without writing them up means the insights die in Dovetail. Writing a memo without citations means the next hire reads opinion, not evidence. Putting the synthesis in Dock fixes all three: the memo is narrative, the rows are citations, and the agent is named. When a new PM joins, they read the memo and click through to the calls. When the founder revisits a decision six months later, the row history shows who approved what. This is the Cloud 2.0 product shape: workspaces that hold decisions, not just documents.
The agent has its own identity, scoped to read Grain and Dovetail and to append to one Notion page. See agent identity for the model, and agent identity lifecycle for how that scope rotates when the agent is retired.
Bob Moesta, interviewed on Lenny Rachitsky's podcast, argues that founders should listen for "struggling moments" rather than feature requests, because demand shows up as a switch, not a wish (Lenny's Newsletter). The Dock surface enforces this: the claim field asks for the trigger, not the ask. Pendo's product analytics framework treats events and event properties as the two raw data types worth tracking (Pendo); the CustomerInsights surface is the qualitative analog, with the call as the event and the tag as the property.
Run your next five customer calls through this loop and read the memo on Monday.
FAQ
Q: Does the agent ever auto-approve insights? No. Confidence scores are visible, but status only changes when the founder clicks approve. Rejected rows stay in the table with a reason field, so the same claim does not get re-proposed next week.
Q: What if Dovetail tags drift over time? The agent reads the current Dovetail taxonomy on every run. When it proposes a new tag, the row is flagged taxonomy-new and routed to the founder for a yes or no before any other rows ship.
Q: Can the memo be exported? Yes. The Notion page is the source of truth and exports as markdown. The Dock rows export as CSV with the Grain and Dovetail pointers intact, so a board update can cite the underlying calls.
Q: How is this different from just using Dovetail's AI summary? Dovetail summarizes clips. Dock holds the founder's approved interpretation of those clips, linked to a narrative memo, with reviewer and timestamp. The two layers compose; they do not replace each other.
